A Low Computational EEG-Based Hand Movements Classification Using a Restricted Boltzmann Machine for Brain-Computer Interface Applications

Hiren Mewada, Miral Desai, Ivan Miguel Pires

A non-invasive brain-computer interface is an innovative approach to a control device without physical execution. Electroen-cephalography (EEG) is the key for these applications. However, classifying EEG signals using fewer computational models is challenging for these applications. This paper classifies hand movement into three classes: left, right, and up. EEG data were acquired from the scalp for three hand movements. A least computational model utilizing a Restricted Boltzmann Machine (RBM) and linear classifier is proposed, which easily fits edge computing devices. The proposed model is evaluated on the power spectrum of the five frequency bands of EEG data. The proposed model succeeded in achieving 96.09%, 90.06%, and 89.53% classification accuracy for training, validation, and test datasets, respectively. The model received 95% and 87% F1 scores in the training and test datasets. Notably, the compact size of the model, i.e., 0.3 MB, shows that the proposed approach offers a computationally efficient approach for real-time applications in EEG-based BCI applications.